The Dead Sea Scrolls, a treasure trove of ancient Jewish manuscripts discovered in 1947 near the shores of the Dead Sea, have long captivated scholars and historians. However, while some of the scrolls remain intact, many fragments have proven difficult or impossible to decipher due to their deteriorated condition.
Now a groundbreaking AI system developed by researchers at Ben-Gurion University of the Negev (BGU) may finally unlock the secrets of these fragmented texts.
‘Embible’ utilizes masked language modeling (MLM) to predict missing text in damaged inscriptions written in Hebrew and Aramaic. To train ‘Embible,’ researchers used 22,144 sentences from the Old Testament, primarily in Hebrew. They first input modern Hebrew data into large language models, then used it to create a model based on ancient Hebrew.
“In the case of a damaged ancient inscription, the parts that are missing might be different,” BGU Prof. Mark Last tells NoCamels. “Sometimes they include one word, sometimes they include a partial word, sometimes they include several words.”
“We worked with biblical text from the Old Testament because in that case, we know the ground truth. So if we randomly mask words or parts of words and try to predict what is missing, we can always check how accurate our prediction was,” he said.
BGU’s ‘Embible’ project was presented at a Wednesday gathering of the European Chapter of the Association for Computational Linguistics in Malta.
